xref: /freebsd/sys/ufs/ffs/ffs_vfsops.c (revision 6f9c8e5b074419423648ffb89b83fd2f257e90b7)
1 /*-
2  * Copyright (c) 1989, 1991, 1993, 1994
3  *	The Regents of the University of California.  All rights reserved.
4  *
5  * Redistribution and use in source and binary forms, with or without
6  * modification, are permitted provided that the following conditions
7  * are met:
8  * 1. Redistributions of source code must retain the above copyright
9  *    notice, this list of conditions and the following disclaimer.
10  * 2. Redistributions in binary form must reproduce the above copyright
11  *    notice, this list of conditions and the following disclaimer in the
12  *    documentation and/or other materials provided with the distribution.
13  * 4. Neither the name of the University nor the names of its contributors
14  *    may be used to endorse or promote products derived from this software
15  *    without specific prior written permission.
16  *
17  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
27  * SUCH DAMAGE.
28  *
29  *	@(#)ffs_vfsops.c	8.31 (Berkeley) 5/20/95
30  */
31 
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
34 
35 #include "opt_quota.h"
36 #include "opt_ufs.h"
37 #include "opt_ffs.h"
38 #include "opt_ddb.h"
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/namei.h>
43 #include <sys/priv.h>
44 #include <sys/proc.h>
45 #include <sys/kernel.h>
46 #include <sys/vnode.h>
47 #include <sys/mount.h>
48 #include <sys/bio.h>
49 #include <sys/buf.h>
50 #include <sys/conf.h>
51 #include <sys/fcntl.h>
52 #include <sys/malloc.h>
53 #include <sys/mutex.h>
54 
55 #include <security/mac/mac_framework.h>
56 
57 #include <ufs/ufs/extattr.h>
58 #include <ufs/ufs/gjournal.h>
59 #include <ufs/ufs/quota.h>
60 #include <ufs/ufs/ufsmount.h>
61 #include <ufs/ufs/inode.h>
62 #include <ufs/ufs/ufs_extern.h>
63 
64 #include <ufs/ffs/fs.h>
65 #include <ufs/ffs/ffs_extern.h>
66 
67 #include <vm/vm.h>
68 #include <vm/uma.h>
69 #include <vm/vm_page.h>
70 
71 #include <geom/geom.h>
72 #include <geom/geom_vfs.h>
73 
74 #include <ddb/ddb.h>
75 
76 static uma_zone_t uma_inode, uma_ufs1, uma_ufs2;
77 
78 static int	ffs_reload(struct mount *, struct thread *);
79 static int	ffs_mountfs(struct vnode *, struct mount *, struct thread *);
80 static void	ffs_oldfscompat_read(struct fs *, struct ufsmount *,
81 		    ufs2_daddr_t);
82 static void	ffs_ifree(struct ufsmount *ump, struct inode *ip);
83 static vfs_init_t ffs_init;
84 static vfs_uninit_t ffs_uninit;
85 static vfs_extattrctl_t ffs_extattrctl;
86 static vfs_cmount_t ffs_cmount;
87 static vfs_unmount_t ffs_unmount;
88 static vfs_mount_t ffs_mount;
89 static vfs_statfs_t ffs_statfs;
90 static vfs_fhtovp_t ffs_fhtovp;
91 static vfs_sync_t ffs_sync;
92 
93 static struct vfsops ufs_vfsops = {
94 	.vfs_extattrctl =	ffs_extattrctl,
95 	.vfs_fhtovp =		ffs_fhtovp,
96 	.vfs_init =		ffs_init,
97 	.vfs_mount =		ffs_mount,
98 	.vfs_cmount =		ffs_cmount,
99 	.vfs_quotactl =		ufs_quotactl,
100 	.vfs_root =		ufs_root,
101 	.vfs_statfs =		ffs_statfs,
102 	.vfs_sync =		ffs_sync,
103 	.vfs_uninit =		ffs_uninit,
104 	.vfs_unmount =		ffs_unmount,
105 	.vfs_vget =		ffs_vget,
106 	.vfs_susp_clean =	process_deferred_inactive,
107 };
108 
109 VFS_SET(ufs_vfsops, ufs, 0);
110 MODULE_VERSION(ufs, 1);
111 
112 static b_strategy_t ffs_geom_strategy;
113 static b_write_t ffs_bufwrite;
114 
115 static struct buf_ops ffs_ops = {
116 	.bop_name =	"FFS",
117 	.bop_write =	ffs_bufwrite,
118 	.bop_strategy =	ffs_geom_strategy,
119 	.bop_sync =	bufsync,
120 #ifdef NO_FFS_SNAPSHOT
121 	.bop_bdflush =	bufbdflush,
122 #else
123 	.bop_bdflush =	ffs_bdflush,
124 #endif
125 };
126 
127 /*
128  * Note that userquota and groupquota options are not currently used
129  * by UFS/FFS code and generally mount(8) does not pass those options
130  * from userland, but they can be passed by loader(8) via
131  * vfs.root.mountfrom.options.
132  */
133 static const char *ffs_opts[] = { "acls", "async", "noatime", "noclusterr",
134     "noclusterw", "noexec", "export", "force", "from", "groupquota",
135     "multilabel", "nfsv4acls", "snapshot", "nosuid", "suiddir", "nosymfollow",
136     "sync", "union", "userquota", NULL };
137 
138 static int
139 ffs_mount(struct mount *mp)
140 {
141 	struct vnode *devvp;
142 	struct thread *td;
143 	struct ufsmount *ump = 0;
144 	struct fs *fs;
145 	int error, flags;
146 	u_int mntorflags;
147 	accmode_t accmode;
148 	struct nameidata ndp;
149 	char *fspec;
150 
151 	td = curthread;
152 	if (vfs_filteropt(mp->mnt_optnew, ffs_opts))
153 		return (EINVAL);
154 	if (uma_inode == NULL) {
155 		uma_inode = uma_zcreate("FFS inode",
156 		    sizeof(struct inode), NULL, NULL, NULL, NULL,
157 		    UMA_ALIGN_PTR, 0);
158 		uma_ufs1 = uma_zcreate("FFS1 dinode",
159 		    sizeof(struct ufs1_dinode), NULL, NULL, NULL, NULL,
160 		    UMA_ALIGN_PTR, 0);
161 		uma_ufs2 = uma_zcreate("FFS2 dinode",
162 		    sizeof(struct ufs2_dinode), NULL, NULL, NULL, NULL,
163 		    UMA_ALIGN_PTR, 0);
164 	}
165 
166 	vfs_deleteopt(mp->mnt_optnew, "groupquota");
167 	vfs_deleteopt(mp->mnt_optnew, "userquota");
168 
169 	fspec = vfs_getopts(mp->mnt_optnew, "from", &error);
170 	if (error)
171 		return (error);
172 
173 	mntorflags = 0;
174 	if (vfs_getopt(mp->mnt_optnew, "acls", NULL, NULL) == 0)
175 		mntorflags |= MNT_ACLS;
176 
177 	if (vfs_getopt(mp->mnt_optnew, "snapshot", NULL, NULL) == 0) {
178 		mntorflags |= MNT_SNAPSHOT;
179 		/*
180 		 * Once we have set the MNT_SNAPSHOT flag, do not
181 		 * persist "snapshot" in the options list.
182 		 */
183 		vfs_deleteopt(mp->mnt_optnew, "snapshot");
184 		vfs_deleteopt(mp->mnt_opt, "snapshot");
185 	}
186 
187 	if (vfs_getopt(mp->mnt_optnew, "nfsv4acls", NULL, NULL) == 0) {
188 		if (mntorflags & MNT_ACLS) {
189 			printf("WARNING: \"acls\" and \"nfsv4acls\" "
190 			    "options are mutually exclusive\n");
191 			return (EINVAL);
192 		}
193 		mntorflags |= MNT_NFS4ACLS;
194 	}
195 
196 	MNT_ILOCK(mp);
197 	mp->mnt_flag |= mntorflags;
198 	MNT_IUNLOCK(mp);
199 	/*
200 	 * If updating, check whether changing from read-only to
201 	 * read/write; if there is no device name, that's all we do.
202 	 */
203 	if (mp->mnt_flag & MNT_UPDATE) {
204 		ump = VFSTOUFS(mp);
205 		fs = ump->um_fs;
206 		devvp = ump->um_devvp;
207 		if (fs->fs_ronly == 0 &&
208 		    vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
209 			/*
210 			 * Flush any dirty data and suspend filesystem.
211 			 */
212 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
213 				return (error);
214 			for (;;) {
215 				vn_finished_write(mp);
216 				if ((error = vfs_write_suspend(mp)) != 0)
217 					return (error);
218 				MNT_ILOCK(mp);
219 				if (mp->mnt_kern_flag & MNTK_SUSPENDED) {
220 					/*
221 					 * Allow the secondary writes
222 					 * to proceed.
223 					 */
224 					mp->mnt_kern_flag &= ~(MNTK_SUSPENDED |
225 					    MNTK_SUSPEND2);
226 					wakeup(&mp->mnt_flag);
227 					MNT_IUNLOCK(mp);
228 					/*
229 					 * Allow the curthread to
230 					 * ignore the suspension to
231 					 * synchronize on-disk state.
232 					 */
233 					td->td_pflags |= TDP_IGNSUSP;
234 					break;
235 				}
236 				MNT_IUNLOCK(mp);
237 				vn_start_write(NULL, &mp, V_WAIT);
238 			}
239 			/*
240 			 * Check for and optionally get rid of files open
241 			 * for writing.
242 			 */
243 			flags = WRITECLOSE;
244 			if (mp->mnt_flag & MNT_FORCE)
245 				flags |= FORCECLOSE;
246 			if (mp->mnt_flag & MNT_SOFTDEP) {
247 				error = softdep_flushfiles(mp, flags, td);
248 			} else {
249 				error = ffs_flushfiles(mp, flags, td);
250 			}
251 			if (error) {
252 				vfs_write_resume(mp);
253 				return (error);
254 			}
255 			if (fs->fs_pendingblocks != 0 ||
256 			    fs->fs_pendinginodes != 0) {
257 				printf("%s: %s: blocks %jd files %d\n",
258 				    fs->fs_fsmnt, "update error",
259 				    (intmax_t)fs->fs_pendingblocks,
260 				    fs->fs_pendinginodes);
261 				fs->fs_pendingblocks = 0;
262 				fs->fs_pendinginodes = 0;
263 			}
264 			if ((fs->fs_flags & (FS_UNCLEAN | FS_NEEDSFSCK)) == 0)
265 				fs->fs_clean = 1;
266 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
267 				fs->fs_ronly = 0;
268 				fs->fs_clean = 0;
269 				vfs_write_resume(mp);
270 				return (error);
271 			}
272 			if (mp->mnt_flag & MNT_SOFTDEP)
273 				softdep_unmount(mp);
274 			DROP_GIANT();
275 			g_topology_lock();
276 			g_access(ump->um_cp, 0, -1, 0);
277 			g_topology_unlock();
278 			PICKUP_GIANT();
279 			fs->fs_ronly = 1;
280 			MNT_ILOCK(mp);
281 			mp->mnt_flag |= MNT_RDONLY;
282 			MNT_IUNLOCK(mp);
283 			/*
284 			 * Allow the writers to note that filesystem
285 			 * is ro now.
286 			 */
287 			vfs_write_resume(mp);
288 		}
289 		if ((mp->mnt_flag & MNT_RELOAD) &&
290 		    (error = ffs_reload(mp, td)) != 0)
291 			return (error);
292 		if (fs->fs_ronly &&
293 		    !vfs_flagopt(mp->mnt_optnew, "ro", NULL, 0)) {
294 			/*
295 			 * If upgrade to read-write by non-root, then verify
296 			 * that user has necessary permissions on the device.
297 			 */
298 			vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
299 			error = VOP_ACCESS(devvp, VREAD | VWRITE,
300 			    td->td_ucred, td);
301 			if (error)
302 				error = priv_check(td, PRIV_VFS_MOUNT_PERM);
303 			if (error) {
304 				VOP_UNLOCK(devvp, 0);
305 				return (error);
306 			}
307 			VOP_UNLOCK(devvp, 0);
308 			fs->fs_flags &= ~FS_UNCLEAN;
309 			if (fs->fs_clean == 0) {
310 				fs->fs_flags |= FS_UNCLEAN;
311 				if ((mp->mnt_flag & MNT_FORCE) ||
312 				    ((fs->fs_flags &
313 				     (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
314 				     (fs->fs_flags & FS_DOSOFTDEP))) {
315 					printf("WARNING: %s was not %s\n",
316 					   fs->fs_fsmnt, "properly dismounted");
317 				} else {
318 					printf(
319 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
320 					    fs->fs_fsmnt);
321 					if (fs->fs_flags & FS_SUJ)
322 						printf(
323 "WARNING: Forced mount will invalidate journal contents\n");
324 					return (EPERM);
325 				}
326 			}
327 			DROP_GIANT();
328 			g_topology_lock();
329 			/*
330 			 * If we're the root device, we may not have an E count
331 			 * yet, get it now.
332 			 */
333 			if (ump->um_cp->ace == 0)
334 				error = g_access(ump->um_cp, 0, 1, 1);
335 			else
336 				error = g_access(ump->um_cp, 0, 1, 0);
337 			g_topology_unlock();
338 			PICKUP_GIANT();
339 			if (error)
340 				return (error);
341 			if ((error = vn_start_write(NULL, &mp, V_WAIT)) != 0)
342 				return (error);
343 			fs->fs_ronly = 0;
344 			MNT_ILOCK(mp);
345 			mp->mnt_flag &= ~MNT_RDONLY;
346 			MNT_IUNLOCK(mp);
347 			fs->fs_mtime = time_second;
348 			/* check to see if we need to start softdep */
349 			if ((fs->fs_flags & FS_DOSOFTDEP) &&
350 			    (error = softdep_mount(devvp, mp, fs, td->td_ucred))){
351 				vn_finished_write(mp);
352 				return (error);
353 			}
354 			fs->fs_clean = 0;
355 			if ((error = ffs_sbupdate(ump, MNT_WAIT, 0)) != 0) {
356 				vn_finished_write(mp);
357 				return (error);
358 			}
359 			if (fs->fs_snapinum[0] != 0)
360 				ffs_snapshot_mount(mp);
361 			vn_finished_write(mp);
362 		}
363 		/*
364 		 * Soft updates is incompatible with "async",
365 		 * so if we are doing softupdates stop the user
366 		 * from setting the async flag in an update.
367 		 * Softdep_mount() clears it in an initial mount
368 		 * or ro->rw remount.
369 		 */
370 		if (mp->mnt_flag & MNT_SOFTDEP) {
371 			/* XXX: Reset too late ? */
372 			MNT_ILOCK(mp);
373 			mp->mnt_flag &= ~MNT_ASYNC;
374 			MNT_IUNLOCK(mp);
375 		}
376 		/*
377 		 * Keep MNT_ACLS flag if it is stored in superblock.
378 		 */
379 		if ((fs->fs_flags & FS_ACLS) != 0) {
380 			/* XXX: Set too late ? */
381 			MNT_ILOCK(mp);
382 			mp->mnt_flag |= MNT_ACLS;
383 			MNT_IUNLOCK(mp);
384 		}
385 
386 		if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
387 			/* XXX: Set too late ? */
388 			MNT_ILOCK(mp);
389 			mp->mnt_flag |= MNT_NFS4ACLS;
390 			MNT_IUNLOCK(mp);
391 		}
392 
393 		/*
394 		 * If this is a snapshot request, take the snapshot.
395 		 */
396 		if (mp->mnt_flag & MNT_SNAPSHOT)
397 			return (ffs_snapshot(mp, fspec));
398 	}
399 
400 	/*
401 	 * Not an update, or updating the name: look up the name
402 	 * and verify that it refers to a sensible disk device.
403 	 */
404 	NDINIT(&ndp, LOOKUP, FOLLOW | LOCKLEAF, UIO_SYSSPACE, fspec, td);
405 	if ((error = namei(&ndp)) != 0)
406 		return (error);
407 	NDFREE(&ndp, NDF_ONLY_PNBUF);
408 	devvp = ndp.ni_vp;
409 	if (!vn_isdisk(devvp, &error)) {
410 		vput(devvp);
411 		return (error);
412 	}
413 
414 	/*
415 	 * If mount by non-root, then verify that user has necessary
416 	 * permissions on the device.
417 	 */
418 	accmode = VREAD;
419 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
420 		accmode |= VWRITE;
421 	error = VOP_ACCESS(devvp, accmode, td->td_ucred, td);
422 	if (error)
423 		error = priv_check(td, PRIV_VFS_MOUNT_PERM);
424 	if (error) {
425 		vput(devvp);
426 		return (error);
427 	}
428 
429 	if (mp->mnt_flag & MNT_UPDATE) {
430 		/*
431 		 * Update only
432 		 *
433 		 * If it's not the same vnode, or at least the same device
434 		 * then it's not correct.
435 		 */
436 
437 		if (devvp->v_rdev != ump->um_devvp->v_rdev)
438 			error = EINVAL;	/* needs translation */
439 		vput(devvp);
440 		if (error)
441 			return (error);
442 	} else {
443 		/*
444 		 * New mount
445 		 *
446 		 * We need the name for the mount point (also used for
447 		 * "last mounted on") copied in. If an error occurs,
448 		 * the mount point is discarded by the upper level code.
449 		 * Note that vfs_mount() populates f_mntonname for us.
450 		 */
451 		if ((error = ffs_mountfs(devvp, mp, td)) != 0) {
452 			vrele(devvp);
453 			return (error);
454 		}
455 	}
456 	vfs_mountedfrom(mp, fspec);
457 	return (0);
458 }
459 
460 /*
461  * Compatibility with old mount system call.
462  */
463 
464 static int
465 ffs_cmount(struct mntarg *ma, void *data, int flags)
466 {
467 	struct ufs_args args;
468 	struct export_args exp;
469 	int error;
470 
471 	if (data == NULL)
472 		return (EINVAL);
473 	error = copyin(data, &args, sizeof args);
474 	if (error)
475 		return (error);
476 	vfs_oexport_conv(&args.export, &exp);
477 
478 	ma = mount_argsu(ma, "from", args.fspec, MAXPATHLEN);
479 	ma = mount_arg(ma, "export", &exp, sizeof(exp));
480 	error = kernel_mount(ma, flags);
481 
482 	return (error);
483 }
484 
485 /*
486  * Reload all incore data for a filesystem (used after running fsck on
487  * the root filesystem and finding things to fix). The filesystem must
488  * be mounted read-only.
489  *
490  * Things to do to update the mount:
491  *	1) invalidate all cached meta-data.
492  *	2) re-read superblock from disk.
493  *	3) re-read summary information from disk.
494  *	4) invalidate all inactive vnodes.
495  *	5) invalidate all cached file data.
496  *	6) re-read inode data for all active vnodes.
497  */
498 static int
499 ffs_reload(struct mount *mp, struct thread *td)
500 {
501 	struct vnode *vp, *mvp, *devvp;
502 	struct inode *ip;
503 	void *space;
504 	struct buf *bp;
505 	struct fs *fs, *newfs;
506 	struct ufsmount *ump;
507 	ufs2_daddr_t sblockloc;
508 	int i, blks, size, error;
509 	int32_t *lp;
510 
511 	if ((mp->mnt_flag & MNT_RDONLY) == 0)
512 		return (EINVAL);
513 	ump = VFSTOUFS(mp);
514 	/*
515 	 * Step 1: invalidate all cached meta-data.
516 	 */
517 	devvp = VFSTOUFS(mp)->um_devvp;
518 	vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
519 	if (vinvalbuf(devvp, 0, 0, 0) != 0)
520 		panic("ffs_reload: dirty1");
521 	VOP_UNLOCK(devvp, 0);
522 
523 	/*
524 	 * Step 2: re-read superblock from disk.
525 	 */
526 	fs = VFSTOUFS(mp)->um_fs;
527 	if ((error = bread(devvp, btodb(fs->fs_sblockloc), fs->fs_sbsize,
528 	    NOCRED, &bp)) != 0)
529 		return (error);
530 	newfs = (struct fs *)bp->b_data;
531 	if ((newfs->fs_magic != FS_UFS1_MAGIC &&
532 	     newfs->fs_magic != FS_UFS2_MAGIC) ||
533 	    newfs->fs_bsize > MAXBSIZE ||
534 	    newfs->fs_bsize < sizeof(struct fs)) {
535 			brelse(bp);
536 			return (EIO);		/* XXX needs translation */
537 	}
538 	/*
539 	 * Copy pointer fields back into superblock before copying in	XXX
540 	 * new superblock. These should really be in the ufsmount.	XXX
541 	 * Note that important parameters (eg fs_ncg) are unchanged.
542 	 */
543 	newfs->fs_csp = fs->fs_csp;
544 	newfs->fs_maxcluster = fs->fs_maxcluster;
545 	newfs->fs_contigdirs = fs->fs_contigdirs;
546 	newfs->fs_active = fs->fs_active;
547 	/* The file system is still read-only. */
548 	newfs->fs_ronly = 1;
549 	sblockloc = fs->fs_sblockloc;
550 	bcopy(newfs, fs, (u_int)fs->fs_sbsize);
551 	brelse(bp);
552 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
553 	ffs_oldfscompat_read(fs, VFSTOUFS(mp), sblockloc);
554 	UFS_LOCK(ump);
555 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
556 		printf("%s: reload pending error: blocks %jd files %d\n",
557 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
558 		    fs->fs_pendinginodes);
559 		fs->fs_pendingblocks = 0;
560 		fs->fs_pendinginodes = 0;
561 	}
562 	UFS_UNLOCK(ump);
563 
564 	/*
565 	 * Step 3: re-read summary information from disk.
566 	 */
567 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
568 	space = fs->fs_csp;
569 	for (i = 0; i < blks; i += fs->fs_frag) {
570 		size = fs->fs_bsize;
571 		if (i + fs->fs_frag > blks)
572 			size = (blks - i) * fs->fs_fsize;
573 		error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
574 		    NOCRED, &bp);
575 		if (error)
576 			return (error);
577 		bcopy(bp->b_data, space, (u_int)size);
578 		space = (char *)space + size;
579 		brelse(bp);
580 	}
581 	/*
582 	 * We no longer know anything about clusters per cylinder group.
583 	 */
584 	if (fs->fs_contigsumsize > 0) {
585 		lp = fs->fs_maxcluster;
586 		for (i = 0; i < fs->fs_ncg; i++)
587 			*lp++ = fs->fs_contigsumsize;
588 	}
589 
590 loop:
591 	MNT_ILOCK(mp);
592 	MNT_VNODE_FOREACH(vp, mp, mvp) {
593 		VI_LOCK(vp);
594 		if (vp->v_iflag & VI_DOOMED) {
595 			VI_UNLOCK(vp);
596 			continue;
597 		}
598 		MNT_IUNLOCK(mp);
599 		/*
600 		 * Step 4: invalidate all cached file data.
601 		 */
602 		if (vget(vp, LK_EXCLUSIVE | LK_INTERLOCK, td)) {
603 			MNT_VNODE_FOREACH_ABORT(mp, mvp);
604 			goto loop;
605 		}
606 		if (vinvalbuf(vp, 0, 0, 0))
607 			panic("ffs_reload: dirty2");
608 		/*
609 		 * Step 5: re-read inode data for all active vnodes.
610 		 */
611 		ip = VTOI(vp);
612 		error =
613 		    bread(devvp, fsbtodb(fs, ino_to_fsba(fs, ip->i_number)),
614 		    (int)fs->fs_bsize, NOCRED, &bp);
615 		if (error) {
616 			VOP_UNLOCK(vp, 0);
617 			vrele(vp);
618 			MNT_VNODE_FOREACH_ABORT(mp, mvp);
619 			return (error);
620 		}
621 		ffs_load_inode(bp, ip, fs, ip->i_number);
622 		ip->i_effnlink = ip->i_nlink;
623 		brelse(bp);
624 		VOP_UNLOCK(vp, 0);
625 		vrele(vp);
626 		MNT_ILOCK(mp);
627 	}
628 	MNT_IUNLOCK(mp);
629 	return (0);
630 }
631 
632 /*
633  * Possible superblock locations ordered from most to least likely.
634  */
635 static int sblock_try[] = SBLOCKSEARCH;
636 
637 /*
638  * Common code for mount and mountroot
639  */
640 static int
641 ffs_mountfs(devvp, mp, td)
642 	struct vnode *devvp;
643 	struct mount *mp;
644 	struct thread *td;
645 {
646 	struct ufsmount *ump;
647 	struct buf *bp;
648 	struct fs *fs;
649 	struct cdev *dev;
650 	void *space;
651 	ufs2_daddr_t sblockloc;
652 	int error, i, blks, size, ronly;
653 	int32_t *lp;
654 	struct ucred *cred;
655 	struct g_consumer *cp;
656 	struct mount *nmp;
657 
658 	bp = NULL;
659 	ump = NULL;
660 	cred = td ? td->td_ucred : NOCRED;
661 	ronly = (mp->mnt_flag & MNT_RDONLY) != 0;
662 
663 	dev = devvp->v_rdev;
664 	dev_ref(dev);
665 	DROP_GIANT();
666 	g_topology_lock();
667 	error = g_vfs_open(devvp, &cp, "ffs", ronly ? 0 : 1);
668 
669 	/*
670 	 * If we are a root mount, drop the E flag so fsck can do its magic.
671 	 * We will pick it up again when we remount R/W.
672 	 */
673 	if (error == 0 && ronly && (mp->mnt_flag & MNT_ROOTFS))
674 		error = g_access(cp, 0, 0, -1);
675 	g_topology_unlock();
676 	PICKUP_GIANT();
677 	VOP_UNLOCK(devvp, 0);
678 	if (error)
679 		goto out;
680 	if (devvp->v_rdev->si_iosize_max != 0)
681 		mp->mnt_iosize_max = devvp->v_rdev->si_iosize_max;
682 	if (mp->mnt_iosize_max > MAXPHYS)
683 		mp->mnt_iosize_max = MAXPHYS;
684 
685 	devvp->v_bufobj.bo_ops = &ffs_ops;
686 
687 	fs = NULL;
688 	sblockloc = 0;
689 	/*
690 	 * Try reading the superblock in each of its possible locations.
691 	 */
692 	for (i = 0; sblock_try[i] != -1; i++) {
693 		if ((SBLOCKSIZE % cp->provider->sectorsize) != 0) {
694 			error = EINVAL;
695 			vfs_mount_error(mp,
696 			    "Invalid sectorsize %d for superblock size %d",
697 			    cp->provider->sectorsize, SBLOCKSIZE);
698 			goto out;
699 		}
700 		if ((error = bread(devvp, btodb(sblock_try[i]), SBLOCKSIZE,
701 		    cred, &bp)) != 0)
702 			goto out;
703 		fs = (struct fs *)bp->b_data;
704 		sblockloc = sblock_try[i];
705 		if ((fs->fs_magic == FS_UFS1_MAGIC ||
706 		     (fs->fs_magic == FS_UFS2_MAGIC &&
707 		      (fs->fs_sblockloc == sblockloc ||
708 		       (fs->fs_old_flags & FS_FLAGS_UPDATED) == 0))) &&
709 		    fs->fs_bsize <= MAXBSIZE &&
710 		    fs->fs_bsize >= sizeof(struct fs))
711 			break;
712 		brelse(bp);
713 		bp = NULL;
714 	}
715 	if (sblock_try[i] == -1) {
716 		error = EINVAL;		/* XXX needs translation */
717 		goto out;
718 	}
719 	fs->fs_fmod = 0;
720 	fs->fs_flags &= ~FS_INDEXDIRS;	/* no support for directory indicies */
721 	fs->fs_flags &= ~FS_UNCLEAN;
722 	if (fs->fs_clean == 0) {
723 		fs->fs_flags |= FS_UNCLEAN;
724 		if (ronly || (mp->mnt_flag & MNT_FORCE) ||
725 		    ((fs->fs_flags & (FS_SUJ | FS_NEEDSFSCK)) == 0 &&
726 		     (fs->fs_flags & FS_DOSOFTDEP))) {
727 			printf("WARNING: %s was not properly dismounted\n",
728 			    fs->fs_fsmnt);
729 		} else {
730 			printf(
731 "WARNING: R/W mount of %s denied.  Filesystem is not clean - run fsck\n",
732 			    fs->fs_fsmnt);
733 			if (fs->fs_flags & FS_SUJ)
734 				printf(
735 "WARNING: Forced mount will invalidate journal contents\n");
736 			error = EPERM;
737 			goto out;
738 		}
739 		if ((fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) &&
740 		    (mp->mnt_flag & MNT_FORCE)) {
741 			printf("%s: lost blocks %jd files %d\n", fs->fs_fsmnt,
742 			    (intmax_t)fs->fs_pendingblocks,
743 			    fs->fs_pendinginodes);
744 			fs->fs_pendingblocks = 0;
745 			fs->fs_pendinginodes = 0;
746 		}
747 	}
748 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
749 		printf("%s: mount pending error: blocks %jd files %d\n",
750 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
751 		    fs->fs_pendinginodes);
752 		fs->fs_pendingblocks = 0;
753 		fs->fs_pendinginodes = 0;
754 	}
755 	if ((fs->fs_flags & FS_GJOURNAL) != 0) {
756 #ifdef UFS_GJOURNAL
757 		/*
758 		 * Get journal provider name.
759 		 */
760 		size = 1024;
761 		mp->mnt_gjprovider = malloc(size, M_UFSMNT, M_WAITOK);
762 		if (g_io_getattr("GJOURNAL::provider", cp, &size,
763 		    mp->mnt_gjprovider) == 0) {
764 			mp->mnt_gjprovider = realloc(mp->mnt_gjprovider, size,
765 			    M_UFSMNT, M_WAITOK);
766 			MNT_ILOCK(mp);
767 			mp->mnt_flag |= MNT_GJOURNAL;
768 			MNT_IUNLOCK(mp);
769 		} else {
770 			printf(
771 "WARNING: %s: GJOURNAL flag on fs but no gjournal provider below\n",
772 			    mp->mnt_stat.f_mntonname);
773 			free(mp->mnt_gjprovider, M_UFSMNT);
774 			mp->mnt_gjprovider = NULL;
775 		}
776 #else
777 		printf(
778 "WARNING: %s: GJOURNAL flag on fs but no UFS_GJOURNAL support\n",
779 		    mp->mnt_stat.f_mntonname);
780 #endif
781 	} else {
782 		mp->mnt_gjprovider = NULL;
783 	}
784 	ump = malloc(sizeof *ump, M_UFSMNT, M_WAITOK | M_ZERO);
785 	ump->um_cp = cp;
786 	ump->um_bo = &devvp->v_bufobj;
787 	ump->um_fs = malloc((u_long)fs->fs_sbsize, M_UFSMNT, M_WAITOK);
788 	if (fs->fs_magic == FS_UFS1_MAGIC) {
789 		ump->um_fstype = UFS1;
790 		ump->um_balloc = ffs_balloc_ufs1;
791 	} else {
792 		ump->um_fstype = UFS2;
793 		ump->um_balloc = ffs_balloc_ufs2;
794 	}
795 	ump->um_blkatoff = ffs_blkatoff;
796 	ump->um_truncate = ffs_truncate;
797 	ump->um_update = ffs_update;
798 	ump->um_valloc = ffs_valloc;
799 	ump->um_vfree = ffs_vfree;
800 	ump->um_ifree = ffs_ifree;
801 	ump->um_rdonly = ffs_rdonly;
802 	ump->um_snapgone = ffs_snapgone;
803 	mtx_init(UFS_MTX(ump), "FFS", "FFS Lock", MTX_DEF);
804 	bcopy(bp->b_data, ump->um_fs, (u_int)fs->fs_sbsize);
805 	if (fs->fs_sbsize < SBLOCKSIZE)
806 		bp->b_flags |= B_INVAL | B_NOCACHE;
807 	brelse(bp);
808 	bp = NULL;
809 	fs = ump->um_fs;
810 	ffs_oldfscompat_read(fs, ump, sblockloc);
811 	fs->fs_ronly = ronly;
812 	size = fs->fs_cssize;
813 	blks = howmany(size, fs->fs_fsize);
814 	if (fs->fs_contigsumsize > 0)
815 		size += fs->fs_ncg * sizeof(int32_t);
816 	size += fs->fs_ncg * sizeof(u_int8_t);
817 	space = malloc((u_long)size, M_UFSMNT, M_WAITOK);
818 	fs->fs_csp = space;
819 	for (i = 0; i < blks; i += fs->fs_frag) {
820 		size = fs->fs_bsize;
821 		if (i + fs->fs_frag > blks)
822 			size = (blks - i) * fs->fs_fsize;
823 		if ((error = bread(devvp, fsbtodb(fs, fs->fs_csaddr + i), size,
824 		    cred, &bp)) != 0) {
825 			free(fs->fs_csp, M_UFSMNT);
826 			goto out;
827 		}
828 		bcopy(bp->b_data, space, (u_int)size);
829 		space = (char *)space + size;
830 		brelse(bp);
831 		bp = NULL;
832 	}
833 	if (fs->fs_contigsumsize > 0) {
834 		fs->fs_maxcluster = lp = space;
835 		for (i = 0; i < fs->fs_ncg; i++)
836 			*lp++ = fs->fs_contigsumsize;
837 		space = lp;
838 	}
839 	size = fs->fs_ncg * sizeof(u_int8_t);
840 	fs->fs_contigdirs = (u_int8_t *)space;
841 	bzero(fs->fs_contigdirs, size);
842 	fs->fs_active = NULL;
843 	mp->mnt_data = ump;
844 	mp->mnt_stat.f_fsid.val[0] = fs->fs_id[0];
845 	mp->mnt_stat.f_fsid.val[1] = fs->fs_id[1];
846 	nmp = NULL;
847 	if (fs->fs_id[0] == 0 || fs->fs_id[1] == 0 ||
848 	    (nmp = vfs_getvfs(&mp->mnt_stat.f_fsid))) {
849 		if (nmp)
850 			vfs_rel(nmp);
851 		vfs_getnewfsid(mp);
852 	}
853 	mp->mnt_maxsymlinklen = fs->fs_maxsymlinklen;
854 	MNT_ILOCK(mp);
855 	mp->mnt_flag |= MNT_LOCAL;
856 	MNT_IUNLOCK(mp);
857 	if ((fs->fs_flags & FS_MULTILABEL) != 0) {
858 #ifdef MAC
859 		MNT_ILOCK(mp);
860 		mp->mnt_flag |= MNT_MULTILABEL;
861 		MNT_IUNLOCK(mp);
862 #else
863 		printf(
864 "WARNING: %s: multilabel flag on fs but no MAC support\n",
865 		    mp->mnt_stat.f_mntonname);
866 #endif
867 	}
868 	if ((fs->fs_flags & FS_ACLS) != 0) {
869 #ifdef UFS_ACL
870 		MNT_ILOCK(mp);
871 
872 		if (mp->mnt_flag & MNT_NFS4ACLS)
873 			printf("WARNING: ACLs flag on fs conflicts with "
874 			    "\"nfsv4acls\" mount option; option ignored\n");
875 		mp->mnt_flag &= ~MNT_NFS4ACLS;
876 		mp->mnt_flag |= MNT_ACLS;
877 
878 		MNT_IUNLOCK(mp);
879 #else
880 		printf("WARNING: %s: ACLs flag on fs but no ACLs support\n",
881 		    mp->mnt_stat.f_mntonname);
882 #endif
883 	}
884 	if ((fs->fs_flags & FS_NFS4ACLS) != 0) {
885 #ifdef UFS_ACL
886 		MNT_ILOCK(mp);
887 
888 		if (mp->mnt_flag & MNT_ACLS)
889 			printf("WARNING: NFSv4 ACLs flag on fs conflicts with "
890 			    "\"acls\" mount option; option ignored\n");
891 		mp->mnt_flag &= ~MNT_ACLS;
892 		mp->mnt_flag |= MNT_NFS4ACLS;
893 
894 		MNT_IUNLOCK(mp);
895 #else
896 		printf(
897 "WARNING: %s: NFSv4 ACLs flag on fs but no ACLs support\n",
898 		    mp->mnt_stat.f_mntonname);
899 #endif
900 	}
901 	if ((fs->fs_flags & FS_TRIM) != 0) {
902 		size = sizeof(int);
903 		if (g_io_getattr("GEOM::candelete", cp, &size,
904 		    &ump->um_candelete) == 0) {
905 			if (!ump->um_candelete)
906 				printf(
907 "WARNING: %s: TRIM flag on fs but disk does not support TRIM\n",
908 				    mp->mnt_stat.f_mntonname);
909 		} else {
910 			printf(
911 "WARNING: %s: TRIM flag on fs but cannot get whether disk supports TRIM\n",
912 			    mp->mnt_stat.f_mntonname);
913 			ump->um_candelete = 0;
914 		}
915 	}
916 
917 	ump->um_mountp = mp;
918 	ump->um_dev = dev;
919 	ump->um_devvp = devvp;
920 	ump->um_nindir = fs->fs_nindir;
921 	ump->um_bptrtodb = fs->fs_fsbtodb;
922 	ump->um_seqinc = fs->fs_frag;
923 	for (i = 0; i < MAXQUOTAS; i++)
924 		ump->um_quotas[i] = NULLVP;
925 #ifdef UFS_EXTATTR
926 	ufs_extattr_uepm_init(&ump->um_extattr);
927 #endif
928 	/*
929 	 * Set FS local "last mounted on" information (NULL pad)
930 	 */
931 	bzero(fs->fs_fsmnt, MAXMNTLEN);
932 	strlcpy(fs->fs_fsmnt, mp->mnt_stat.f_mntonname, MAXMNTLEN);
933 	mp->mnt_stat.f_iosize = fs->fs_bsize;
934 
935 	if( mp->mnt_flag & MNT_ROOTFS) {
936 		/*
937 		 * Root mount; update timestamp in mount structure.
938 		 * this will be used by the common root mount code
939 		 * to update the system clock.
940 		 */
941 		mp->mnt_time = fs->fs_time;
942 	}
943 
944 	if (ronly == 0) {
945 		fs->fs_mtime = time_second;
946 		if ((fs->fs_flags & FS_DOSOFTDEP) &&
947 		    (error = softdep_mount(devvp, mp, fs, cred)) != 0) {
948 			free(fs->fs_csp, M_UFSMNT);
949 			ffs_flushfiles(mp, FORCECLOSE, td);
950 			goto out;
951 		}
952 		if (fs->fs_snapinum[0] != 0)
953 			ffs_snapshot_mount(mp);
954 		fs->fs_fmod = 1;
955 		fs->fs_clean = 0;
956 		(void) ffs_sbupdate(ump, MNT_WAIT, 0);
957 	}
958 	/*
959 	 * Initialize filesystem stat information in mount struct.
960 	 */
961 	MNT_ILOCK(mp);
962 	mp->mnt_kern_flag |= MNTK_MPSAFE | MNTK_LOOKUP_SHARED |
963 	    MNTK_EXTENDED_SHARED;
964 	MNT_IUNLOCK(mp);
965 #ifdef UFS_EXTATTR
966 #ifdef UFS_EXTATTR_AUTOSTART
967 	/*
968 	 *
969 	 * Auto-starting does the following:
970 	 *	- check for /.attribute in the fs, and extattr_start if so
971 	 *	- for each file in .attribute, enable that file with
972 	 * 	  an attribute of the same name.
973 	 * Not clear how to report errors -- probably eat them.
974 	 * This would all happen while the filesystem was busy/not
975 	 * available, so would effectively be "atomic".
976 	 */
977 	(void) ufs_extattr_autostart(mp, td);
978 #endif /* !UFS_EXTATTR_AUTOSTART */
979 #endif /* !UFS_EXTATTR */
980 	return (0);
981 out:
982 	if (bp)
983 		brelse(bp);
984 	if (cp != NULL) {
985 		DROP_GIANT();
986 		g_topology_lock();
987 		g_vfs_close(cp);
988 		g_topology_unlock();
989 		PICKUP_GIANT();
990 	}
991 	if (ump) {
992 		mtx_destroy(UFS_MTX(ump));
993 		if (mp->mnt_gjprovider != NULL) {
994 			free(mp->mnt_gjprovider, M_UFSMNT);
995 			mp->mnt_gjprovider = NULL;
996 		}
997 		free(ump->um_fs, M_UFSMNT);
998 		free(ump, M_UFSMNT);
999 		mp->mnt_data = NULL;
1000 	}
1001 	dev_rel(dev);
1002 	return (error);
1003 }
1004 
1005 #include <sys/sysctl.h>
1006 static int bigcgs = 0;
1007 SYSCTL_INT(_debug, OID_AUTO, bigcgs, CTLFLAG_RW, &bigcgs, 0, "");
1008 
1009 /*
1010  * Sanity checks for loading old filesystem superblocks.
1011  * See ffs_oldfscompat_write below for unwound actions.
1012  *
1013  * XXX - Parts get retired eventually.
1014  * Unfortunately new bits get added.
1015  */
1016 static void
1017 ffs_oldfscompat_read(fs, ump, sblockloc)
1018 	struct fs *fs;
1019 	struct ufsmount *ump;
1020 	ufs2_daddr_t sblockloc;
1021 {
1022 	off_t maxfilesize;
1023 
1024 	/*
1025 	 * If not yet done, update fs_flags location and value of fs_sblockloc.
1026 	 */
1027 	if ((fs->fs_old_flags & FS_FLAGS_UPDATED) == 0) {
1028 		fs->fs_flags = fs->fs_old_flags;
1029 		fs->fs_old_flags |= FS_FLAGS_UPDATED;
1030 		fs->fs_sblockloc = sblockloc;
1031 	}
1032 	/*
1033 	 * If not yet done, update UFS1 superblock with new wider fields.
1034 	 */
1035 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_maxbsize != fs->fs_bsize) {
1036 		fs->fs_maxbsize = fs->fs_bsize;
1037 		fs->fs_time = fs->fs_old_time;
1038 		fs->fs_size = fs->fs_old_size;
1039 		fs->fs_dsize = fs->fs_old_dsize;
1040 		fs->fs_csaddr = fs->fs_old_csaddr;
1041 		fs->fs_cstotal.cs_ndir = fs->fs_old_cstotal.cs_ndir;
1042 		fs->fs_cstotal.cs_nbfree = fs->fs_old_cstotal.cs_nbfree;
1043 		fs->fs_cstotal.cs_nifree = fs->fs_old_cstotal.cs_nifree;
1044 		fs->fs_cstotal.cs_nffree = fs->fs_old_cstotal.cs_nffree;
1045 	}
1046 	if (fs->fs_magic == FS_UFS1_MAGIC &&
1047 	    fs->fs_old_inodefmt < FS_44INODEFMT) {
1048 		fs->fs_maxfilesize = ((uint64_t)1 << 31) - 1;
1049 		fs->fs_qbmask = ~fs->fs_bmask;
1050 		fs->fs_qfmask = ~fs->fs_fmask;
1051 	}
1052 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1053 		ump->um_savedmaxfilesize = fs->fs_maxfilesize;
1054 		maxfilesize = (uint64_t)0x80000000 * fs->fs_bsize - 1;
1055 		if (fs->fs_maxfilesize > maxfilesize)
1056 			fs->fs_maxfilesize = maxfilesize;
1057 	}
1058 	/* Compatibility for old filesystems */
1059 	if (fs->fs_avgfilesize <= 0)
1060 		fs->fs_avgfilesize = AVFILESIZ;
1061 	if (fs->fs_avgfpdir <= 0)
1062 		fs->fs_avgfpdir = AFPDIR;
1063 	if (bigcgs) {
1064 		fs->fs_save_cgsize = fs->fs_cgsize;
1065 		fs->fs_cgsize = fs->fs_bsize;
1066 	}
1067 }
1068 
1069 /*
1070  * Unwinding superblock updates for old filesystems.
1071  * See ffs_oldfscompat_read above for details.
1072  *
1073  * XXX - Parts get retired eventually.
1074  * Unfortunately new bits get added.
1075  */
1076 void
1077 ffs_oldfscompat_write(fs, ump)
1078 	struct fs *fs;
1079 	struct ufsmount *ump;
1080 {
1081 
1082 	/*
1083 	 * Copy back UFS2 updated fields that UFS1 inspects.
1084 	 */
1085 	if (fs->fs_magic == FS_UFS1_MAGIC) {
1086 		fs->fs_old_time = fs->fs_time;
1087 		fs->fs_old_cstotal.cs_ndir = fs->fs_cstotal.cs_ndir;
1088 		fs->fs_old_cstotal.cs_nbfree = fs->fs_cstotal.cs_nbfree;
1089 		fs->fs_old_cstotal.cs_nifree = fs->fs_cstotal.cs_nifree;
1090 		fs->fs_old_cstotal.cs_nffree = fs->fs_cstotal.cs_nffree;
1091 		fs->fs_maxfilesize = ump->um_savedmaxfilesize;
1092 	}
1093 	if (bigcgs) {
1094 		fs->fs_cgsize = fs->fs_save_cgsize;
1095 		fs->fs_save_cgsize = 0;
1096 	}
1097 }
1098 
1099 /*
1100  * unmount system call
1101  */
1102 static int
1103 ffs_unmount(mp, mntflags)
1104 	struct mount *mp;
1105 	int mntflags;
1106 {
1107 	struct thread *td;
1108 	struct ufsmount *ump = VFSTOUFS(mp);
1109 	struct fs *fs;
1110 	int error, flags, susp;
1111 #ifdef UFS_EXTATTR
1112 	int e_restart;
1113 #endif
1114 
1115 	flags = 0;
1116 	td = curthread;
1117 	fs = ump->um_fs;
1118 	if (mntflags & MNT_FORCE) {
1119 		flags |= FORCECLOSE;
1120 		susp = fs->fs_ronly != 0;
1121 	} else
1122 		susp = 0;
1123 #ifdef UFS_EXTATTR
1124 	if ((error = ufs_extattr_stop(mp, td))) {
1125 		if (error != EOPNOTSUPP)
1126 			printf("ffs_unmount: ufs_extattr_stop returned %d\n",
1127 			    error);
1128 		e_restart = 0;
1129 	} else {
1130 		ufs_extattr_uepm_destroy(&ump->um_extattr);
1131 		e_restart = 1;
1132 	}
1133 #endif
1134 	if (susp) {
1135 		/*
1136 		 * dounmount already called vn_start_write().
1137 		 */
1138 		for (;;) {
1139 			vn_finished_write(mp);
1140 			if ((error = vfs_write_suspend(mp)) != 0)
1141 				return (error);
1142 			MNT_ILOCK(mp);
1143 			if (mp->mnt_kern_flag & MNTK_SUSPENDED) {
1144 				mp->mnt_kern_flag &= ~(MNTK_SUSPENDED |
1145 				    MNTK_SUSPEND2);
1146 				wakeup(&mp->mnt_flag);
1147 				MNT_IUNLOCK(mp);
1148 				td->td_pflags |= TDP_IGNSUSP;
1149 				break;
1150 			}
1151 			MNT_IUNLOCK(mp);
1152 			vn_start_write(NULL, &mp, V_WAIT);
1153 		}
1154 	}
1155 	if (mp->mnt_flag & MNT_SOFTDEP)
1156 		error = softdep_flushfiles(mp, flags, td);
1157 	else
1158 		error = ffs_flushfiles(mp, flags, td);
1159 	if (error != 0 && error != ENXIO)
1160 		goto fail;
1161 
1162 	UFS_LOCK(ump);
1163 	if (fs->fs_pendingblocks != 0 || fs->fs_pendinginodes != 0) {
1164 		printf("%s: unmount pending error: blocks %jd files %d\n",
1165 		    fs->fs_fsmnt, (intmax_t)fs->fs_pendingblocks,
1166 		    fs->fs_pendinginodes);
1167 		fs->fs_pendingblocks = 0;
1168 		fs->fs_pendinginodes = 0;
1169 	}
1170 	UFS_UNLOCK(ump);
1171 	softdep_unmount(mp);
1172 	if (fs->fs_ronly == 0) {
1173 		fs->fs_clean = fs->fs_flags & (FS_UNCLEAN|FS_NEEDSFSCK) ? 0 : 1;
1174 		error = ffs_sbupdate(ump, MNT_WAIT, 0);
1175 		if (error && error != ENXIO) {
1176 			fs->fs_clean = 0;
1177 			goto fail;
1178 		}
1179 	}
1180 	if (susp) {
1181 		vfs_write_resume(mp);
1182 		vn_start_write(NULL, &mp, V_WAIT);
1183 	}
1184 	DROP_GIANT();
1185 	g_topology_lock();
1186 	g_vfs_close(ump->um_cp);
1187 	g_topology_unlock();
1188 	PICKUP_GIANT();
1189 	vrele(ump->um_devvp);
1190 	dev_rel(ump->um_dev);
1191 	mtx_destroy(UFS_MTX(ump));
1192 	if (mp->mnt_gjprovider != NULL) {
1193 		free(mp->mnt_gjprovider, M_UFSMNT);
1194 		mp->mnt_gjprovider = NULL;
1195 	}
1196 	free(fs->fs_csp, M_UFSMNT);
1197 	free(fs, M_UFSMNT);
1198 	free(ump, M_UFSMNT);
1199 	mp->mnt_data = NULL;
1200 	MNT_ILOCK(mp);
1201 	mp->mnt_flag &= ~MNT_LOCAL;
1202 	MNT_IUNLOCK(mp);
1203 	return (error);
1204 
1205 fail:
1206 	if (susp) {
1207 		vfs_write_resume(mp);
1208 		vn_start_write(NULL, &mp, V_WAIT);
1209 	}
1210 #ifdef UFS_EXTATTR
1211 	if (e_restart) {
1212 		ufs_extattr_uepm_init(&ump->um_extattr);
1213 #ifdef UFS_EXTATTR_AUTOSTART
1214 		(void) ufs_extattr_autostart(mp, td);
1215 #endif
1216 	}
1217 #endif
1218 
1219 	return (error);
1220 }
1221 
1222 /*
1223  * Flush out all the files in a filesystem.
1224  */
1225 int
1226 ffs_flushfiles(mp, flags, td)
1227 	struct mount *mp;
1228 	int flags;
1229 	struct thread *td;
1230 {
1231 	struct ufsmount *ump;
1232 	int error;
1233 
1234 	ump = VFSTOUFS(mp);
1235 #ifdef QUOTA
1236 	if (mp->mnt_flag & MNT_QUOTA) {
1237 		int i;
1238 		error = vflush(mp, 0, SKIPSYSTEM|flags, td);
1239 		if (error)
1240 			return (error);
1241 		for (i = 0; i < MAXQUOTAS; i++) {
1242 			quotaoff(td, mp, i);
1243 		}
1244 		/*
1245 		 * Here we fall through to vflush again to ensure
1246 		 * that we have gotten rid of all the system vnodes.
1247 		 */
1248 	}
1249 #endif
1250 	ASSERT_VOP_LOCKED(ump->um_devvp, "ffs_flushfiles");
1251 	if (ump->um_devvp->v_vflag & VV_COPYONWRITE) {
1252 		if ((error = vflush(mp, 0, SKIPSYSTEM | flags, td)) != 0)
1253 			return (error);
1254 		ffs_snapshot_unmount(mp);
1255 		flags |= FORCECLOSE;
1256 		/*
1257 		 * Here we fall through to vflush again to ensure
1258 		 * that we have gotten rid of all the system vnodes.
1259 		 */
1260 	}
1261         /*
1262 	 * Flush all the files.
1263 	 */
1264 	if ((error = vflush(mp, 0, flags, td)) != 0)
1265 		return (error);
1266 	/*
1267 	 * Flush filesystem metadata.
1268 	 */
1269 	vn_lock(ump->um_devvp, LK_EXCLUSIVE | LK_RETRY);
1270 	error = VOP_FSYNC(ump->um_devvp, MNT_WAIT, td);
1271 	VOP_UNLOCK(ump->um_devvp, 0);
1272 	return (error);
1273 }
1274 
1275 /*
1276  * Get filesystem statistics.
1277  */
1278 static int
1279 ffs_statfs(mp, sbp)
1280 	struct mount *mp;
1281 	struct statfs *sbp;
1282 {
1283 	struct ufsmount *ump;
1284 	struct fs *fs;
1285 
1286 	ump = VFSTOUFS(mp);
1287 	fs = ump->um_fs;
1288 	if (fs->fs_magic != FS_UFS1_MAGIC && fs->fs_magic != FS_UFS2_MAGIC)
1289 		panic("ffs_statfs");
1290 	sbp->f_version = STATFS_VERSION;
1291 	sbp->f_bsize = fs->fs_fsize;
1292 	sbp->f_iosize = fs->fs_bsize;
1293 	sbp->f_blocks = fs->fs_dsize;
1294 	UFS_LOCK(ump);
1295 	sbp->f_bfree = fs->fs_cstotal.cs_nbfree * fs->fs_frag +
1296 	    fs->fs_cstotal.cs_nffree + dbtofsb(fs, fs->fs_pendingblocks);
1297 	sbp->f_bavail = freespace(fs, fs->fs_minfree) +
1298 	    dbtofsb(fs, fs->fs_pendingblocks);
1299 	sbp->f_files =  fs->fs_ncg * fs->fs_ipg - ROOTINO;
1300 	sbp->f_ffree = fs->fs_cstotal.cs_nifree + fs->fs_pendinginodes;
1301 	UFS_UNLOCK(ump);
1302 	sbp->f_namemax = NAME_MAX;
1303 	return (0);
1304 }
1305 
1306 /*
1307  * Go through the disk queues to initiate sandbagged IO;
1308  * go through the inodes to write those that have been modified;
1309  * initiate the writing of the super block if it has been modified.
1310  *
1311  * Note: we are always called with the filesystem marked `MPBUSY'.
1312  */
1313 static int
1314 ffs_sync(mp, waitfor)
1315 	struct mount *mp;
1316 	int waitfor;
1317 {
1318 	struct vnode *mvp, *vp, *devvp;
1319 	struct thread *td;
1320 	struct inode *ip;
1321 	struct ufsmount *ump = VFSTOUFS(mp);
1322 	struct fs *fs;
1323 	int error, count, wait, lockreq, allerror = 0;
1324 	int suspend;
1325 	int suspended;
1326 	int secondary_writes;
1327 	int secondary_accwrites;
1328 	int softdep_deps;
1329 	int softdep_accdeps;
1330 	struct bufobj *bo;
1331 
1332 	td = curthread;
1333 	fs = ump->um_fs;
1334 	if (fs->fs_fmod != 0 && fs->fs_ronly != 0) {		/* XXX */
1335 		printf("fs = %s\n", fs->fs_fsmnt);
1336 		panic("ffs_sync: rofs mod");
1337 	}
1338 	/*
1339 	 * Write back each (modified) inode.
1340 	 */
1341 	wait = 0;
1342 	suspend = 0;
1343 	suspended = 0;
1344 	lockreq = LK_EXCLUSIVE | LK_NOWAIT;
1345 	if (waitfor == MNT_SUSPEND) {
1346 		suspend = 1;
1347 		waitfor = MNT_WAIT;
1348 	}
1349 	if (waitfor == MNT_WAIT) {
1350 		wait = 1;
1351 		lockreq = LK_EXCLUSIVE;
1352 	}
1353 	lockreq |= LK_INTERLOCK | LK_SLEEPFAIL;
1354 	MNT_ILOCK(mp);
1355 loop:
1356 	/* Grab snapshot of secondary write counts */
1357 	secondary_writes = mp->mnt_secondary_writes;
1358 	secondary_accwrites = mp->mnt_secondary_accwrites;
1359 
1360 	/* Grab snapshot of softdep dependency counts */
1361 	MNT_IUNLOCK(mp);
1362 	softdep_get_depcounts(mp, &softdep_deps, &softdep_accdeps);
1363 	MNT_ILOCK(mp);
1364 
1365 	MNT_VNODE_FOREACH(vp, mp, mvp) {
1366 		/*
1367 		 * Depend on the mntvnode_slock to keep things stable enough
1368 		 * for a quick test.  Since there might be hundreds of
1369 		 * thousands of vnodes, we cannot afford even a subroutine
1370 		 * call unless there's a good chance that we have work to do.
1371 		 */
1372 		VI_LOCK(vp);
1373 		if (vp->v_iflag & VI_DOOMED) {
1374 			VI_UNLOCK(vp);
1375 			continue;
1376 		}
1377 		ip = VTOI(vp);
1378 		if (vp->v_type == VNON || ((ip->i_flag &
1379 		    (IN_ACCESS | IN_CHANGE | IN_MODIFIED | IN_UPDATE)) == 0 &&
1380 		    vp->v_bufobj.bo_dirty.bv_cnt == 0)) {
1381 			VI_UNLOCK(vp);
1382 			continue;
1383 		}
1384 		MNT_IUNLOCK(mp);
1385 		if ((error = vget(vp, lockreq, td)) != 0) {
1386 			MNT_ILOCK(mp);
1387 			if (error == ENOENT || error == ENOLCK) {
1388 				MNT_VNODE_FOREACH_ABORT_ILOCKED(mp, mvp);
1389 				goto loop;
1390 			}
1391 			continue;
1392 		}
1393 		if ((error = ffs_syncvnode(vp, waitfor)) != 0)
1394 			allerror = error;
1395 		vput(vp);
1396 		MNT_ILOCK(mp);
1397 	}
1398 	MNT_IUNLOCK(mp);
1399 	/*
1400 	 * Force stale filesystem control information to be flushed.
1401 	 */
1402 	if (waitfor == MNT_WAIT) {
1403 		if ((error = softdep_flushworklist(ump->um_mountp, &count, td)))
1404 			allerror = error;
1405 		/* Flushed work items may create new vnodes to clean */
1406 		if (allerror == 0 && count) {
1407 			MNT_ILOCK(mp);
1408 			goto loop;
1409 		}
1410 	}
1411 #ifdef QUOTA
1412 	qsync(mp);
1413 #endif
1414 	devvp = ump->um_devvp;
1415 	bo = &devvp->v_bufobj;
1416 	BO_LOCK(bo);
1417 	if (waitfor != MNT_LAZY &&
1418 	    (bo->bo_numoutput > 0 || bo->bo_dirty.bv_cnt > 0)) {
1419 		BO_UNLOCK(bo);
1420 		vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY);
1421 		if ((error = VOP_FSYNC(devvp, waitfor, td)) != 0)
1422 			allerror = error;
1423 		VOP_UNLOCK(devvp, 0);
1424 		if (allerror == 0 && waitfor == MNT_WAIT) {
1425 			MNT_ILOCK(mp);
1426 			goto loop;
1427 		}
1428 	} else if (suspend != 0) {
1429 		if (softdep_check_suspend(mp,
1430 					  devvp,
1431 					  softdep_deps,
1432 					  softdep_accdeps,
1433 					  secondary_writes,
1434 					  secondary_accwrites) != 0)
1435 			goto loop;	/* More work needed */
1436 		mtx_assert(MNT_MTX(mp), MA_OWNED);
1437 		mp->mnt_kern_flag |= MNTK_SUSPEND2 | MNTK_SUSPENDED;
1438 		MNT_IUNLOCK(mp);
1439 		suspended = 1;
1440 	} else
1441 		BO_UNLOCK(bo);
1442 	/*
1443 	 * Write back modified superblock.
1444 	 */
1445 	if (fs->fs_fmod != 0 &&
1446 	    (error = ffs_sbupdate(ump, waitfor, suspended)) != 0)
1447 		allerror = error;
1448 	return (allerror);
1449 }
1450 
1451 int
1452 ffs_vget(mp, ino, flags, vpp)
1453 	struct mount *mp;
1454 	ino_t ino;
1455 	int flags;
1456 	struct vnode **vpp;
1457 {
1458 	return (ffs_vgetf(mp, ino, flags, vpp, 0));
1459 }
1460 
1461 int
1462 ffs_vgetf(mp, ino, flags, vpp, ffs_flags)
1463 	struct mount *mp;
1464 	ino_t ino;
1465 	int flags;
1466 	struct vnode **vpp;
1467 	int ffs_flags;
1468 {
1469 	struct fs *fs;
1470 	struct inode *ip;
1471 	struct ufsmount *ump;
1472 	struct buf *bp;
1473 	struct vnode *vp;
1474 	struct cdev *dev;
1475 	int error;
1476 
1477 	error = vfs_hash_get(mp, ino, flags, curthread, vpp, NULL, NULL);
1478 	if (error || *vpp != NULL)
1479 		return (error);
1480 
1481 	/*
1482 	 * We must promote to an exclusive lock for vnode creation.  This
1483 	 * can happen if lookup is passed LOCKSHARED.
1484  	 */
1485 	if ((flags & LK_TYPE_MASK) == LK_SHARED) {
1486 		flags &= ~LK_TYPE_MASK;
1487 		flags |= LK_EXCLUSIVE;
1488 	}
1489 
1490 	/*
1491 	 * We do not lock vnode creation as it is believed to be too
1492 	 * expensive for such rare case as simultaneous creation of vnode
1493 	 * for same ino by different processes. We just allow them to race
1494 	 * and check later to decide who wins. Let the race begin!
1495 	 */
1496 
1497 	ump = VFSTOUFS(mp);
1498 	dev = ump->um_dev;
1499 	fs = ump->um_fs;
1500 
1501 	/*
1502 	 * If this malloc() is performed after the getnewvnode()
1503 	 * it might block, leaving a vnode with a NULL v_data to be
1504 	 * found by ffs_sync() if a sync happens to fire right then,
1505 	 * which will cause a panic because ffs_sync() blindly
1506 	 * dereferences vp->v_data (as well it should).
1507 	 */
1508 	ip = uma_zalloc(uma_inode, M_WAITOK | M_ZERO);
1509 
1510 	/* Allocate a new vnode/inode. */
1511 	if (fs->fs_magic == FS_UFS1_MAGIC)
1512 		error = getnewvnode("ufs", mp, &ffs_vnodeops1, &vp);
1513 	else
1514 		error = getnewvnode("ufs", mp, &ffs_vnodeops2, &vp);
1515 	if (error) {
1516 		*vpp = NULL;
1517 		uma_zfree(uma_inode, ip);
1518 		return (error);
1519 	}
1520 	/*
1521 	 * FFS supports recursive locking.
1522 	 */
1523 	lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
1524 	VN_LOCK_AREC(vp);
1525 	vp->v_data = ip;
1526 	vp->v_bufobj.bo_bsize = fs->fs_bsize;
1527 	ip->i_vnode = vp;
1528 	ip->i_ump = ump;
1529 	ip->i_fs = fs;
1530 	ip->i_dev = dev;
1531 	ip->i_number = ino;
1532 	ip->i_ea_refs = 0;
1533 #ifdef QUOTA
1534 	{
1535 		int i;
1536 		for (i = 0; i < MAXQUOTAS; i++)
1537 			ip->i_dquot[i] = NODQUOT;
1538 	}
1539 #endif
1540 
1541 	if (ffs_flags & FFSV_FORCEINSMQ)
1542 		vp->v_vflag |= VV_FORCEINSMQ;
1543 	error = insmntque(vp, mp);
1544 	if (error != 0) {
1545 		uma_zfree(uma_inode, ip);
1546 		*vpp = NULL;
1547 		return (error);
1548 	}
1549 	vp->v_vflag &= ~VV_FORCEINSMQ;
1550 	error = vfs_hash_insert(vp, ino, flags, curthread, vpp, NULL, NULL);
1551 	if (error || *vpp != NULL)
1552 		return (error);
1553 
1554 	/* Read in the disk contents for the inode, copy into the inode. */
1555 	error = bread(ump->um_devvp, fsbtodb(fs, ino_to_fsba(fs, ino)),
1556 	    (int)fs->fs_bsize, NOCRED, &bp);
1557 	if (error) {
1558 		/*
1559 		 * The inode does not contain anything useful, so it would
1560 		 * be misleading to leave it on its hash chain. With mode
1561 		 * still zero, it will be unlinked and returned to the free
1562 		 * list by vput().
1563 		 */
1564 		brelse(bp);
1565 		vput(vp);
1566 		*vpp = NULL;
1567 		return (error);
1568 	}
1569 	if (ip->i_ump->um_fstype == UFS1)
1570 		ip->i_din1 = uma_zalloc(uma_ufs1, M_WAITOK);
1571 	else
1572 		ip->i_din2 = uma_zalloc(uma_ufs2, M_WAITOK);
1573 	ffs_load_inode(bp, ip, fs, ino);
1574 	if (DOINGSOFTDEP(vp))
1575 		softdep_load_inodeblock(ip);
1576 	else
1577 		ip->i_effnlink = ip->i_nlink;
1578 	bqrelse(bp);
1579 
1580 	/*
1581 	 * Initialize the vnode from the inode, check for aliases.
1582 	 * Note that the underlying vnode may have changed.
1583 	 */
1584 	if (ip->i_ump->um_fstype == UFS1)
1585 		error = ufs_vinit(mp, &ffs_fifoops1, &vp);
1586 	else
1587 		error = ufs_vinit(mp, &ffs_fifoops2, &vp);
1588 	if (error) {
1589 		vput(vp);
1590 		*vpp = NULL;
1591 		return (error);
1592 	}
1593 
1594 	/*
1595 	 * Finish inode initialization.
1596 	 */
1597 	if (vp->v_type != VFIFO) {
1598 		/* FFS supports shared locking for all files except fifos. */
1599 		VN_LOCK_ASHARE(vp);
1600 	}
1601 
1602 	/*
1603 	 * Set up a generation number for this inode if it does not
1604 	 * already have one. This should only happen on old filesystems.
1605 	 */
1606 	if (ip->i_gen == 0) {
1607 		ip->i_gen = arc4random() / 2 + 1;
1608 		if ((vp->v_mount->mnt_flag & MNT_RDONLY) == 0) {
1609 			ip->i_flag |= IN_MODIFIED;
1610 			DIP_SET(ip, i_gen, ip->i_gen);
1611 		}
1612 	}
1613 #ifdef MAC
1614 	if ((mp->mnt_flag & MNT_MULTILABEL) && ip->i_mode) {
1615 		/*
1616 		 * If this vnode is already allocated, and we're running
1617 		 * multi-label, attempt to perform a label association
1618 		 * from the extended attributes on the inode.
1619 		 */
1620 		error = mac_vnode_associate_extattr(mp, vp);
1621 		if (error) {
1622 			/* ufs_inactive will release ip->i_devvp ref. */
1623 			vput(vp);
1624 			*vpp = NULL;
1625 			return (error);
1626 		}
1627 	}
1628 #endif
1629 
1630 	*vpp = vp;
1631 	return (0);
1632 }
1633 
1634 /*
1635  * File handle to vnode
1636  *
1637  * Have to be really careful about stale file handles:
1638  * - check that the inode number is valid
1639  * - call ffs_vget() to get the locked inode
1640  * - check for an unallocated inode (i_mode == 0)
1641  * - check that the given client host has export rights and return
1642  *   those rights via. exflagsp and credanonp
1643  */
1644 static int
1645 ffs_fhtovp(mp, fhp, flags, vpp)
1646 	struct mount *mp;
1647 	struct fid *fhp;
1648 	int flags;
1649 	struct vnode **vpp;
1650 {
1651 	struct ufid *ufhp;
1652 	struct fs *fs;
1653 
1654 	ufhp = (struct ufid *)fhp;
1655 	fs = VFSTOUFS(mp)->um_fs;
1656 	if (ufhp->ufid_ino < ROOTINO ||
1657 	    ufhp->ufid_ino >= fs->fs_ncg * fs->fs_ipg)
1658 		return (ESTALE);
1659 	return (ufs_fhtovp(mp, ufhp, flags, vpp));
1660 }
1661 
1662 /*
1663  * Initialize the filesystem.
1664  */
1665 static int
1666 ffs_init(vfsp)
1667 	struct vfsconf *vfsp;
1668 {
1669 
1670 	softdep_initialize();
1671 	return (ufs_init(vfsp));
1672 }
1673 
1674 /*
1675  * Undo the work of ffs_init().
1676  */
1677 static int
1678 ffs_uninit(vfsp)
1679 	struct vfsconf *vfsp;
1680 {
1681 	int ret;
1682 
1683 	ret = ufs_uninit(vfsp);
1684 	softdep_uninitialize();
1685 	return (ret);
1686 }
1687 
1688 /*
1689  * Write a superblock and associated information back to disk.
1690  */
1691 int
1692 ffs_sbupdate(mp, waitfor, suspended)
1693 	struct ufsmount *mp;
1694 	int waitfor;
1695 	int suspended;
1696 {
1697 	struct fs *fs = mp->um_fs;
1698 	struct buf *sbbp;
1699 	struct buf *bp;
1700 	int blks;
1701 	void *space;
1702 	int i, size, error, allerror = 0;
1703 
1704 	if (fs->fs_ronly == 1 &&
1705 	    (mp->um_mountp->mnt_flag & (MNT_RDONLY | MNT_UPDATE)) !=
1706 	    (MNT_RDONLY | MNT_UPDATE))
1707 		panic("ffs_sbupdate: write read-only filesystem");
1708 	/*
1709 	 * We use the superblock's buf to serialize calls to ffs_sbupdate().
1710 	 */
1711 	sbbp = getblk(mp->um_devvp, btodb(fs->fs_sblockloc), (int)fs->fs_sbsize,
1712 	    0, 0, 0);
1713 	/*
1714 	 * First write back the summary information.
1715 	 */
1716 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1717 	space = fs->fs_csp;
1718 	for (i = 0; i < blks; i += fs->fs_frag) {
1719 		size = fs->fs_bsize;
1720 		if (i + fs->fs_frag > blks)
1721 			size = (blks - i) * fs->fs_fsize;
1722 		bp = getblk(mp->um_devvp, fsbtodb(fs, fs->fs_csaddr + i),
1723 		    size, 0, 0, 0);
1724 		bcopy(space, bp->b_data, (u_int)size);
1725 		space = (char *)space + size;
1726 		if (suspended)
1727 			bp->b_flags |= B_VALIDSUSPWRT;
1728 		if (waitfor != MNT_WAIT)
1729 			bawrite(bp);
1730 		else if ((error = bwrite(bp)) != 0)
1731 			allerror = error;
1732 	}
1733 	/*
1734 	 * Now write back the superblock itself. If any errors occurred
1735 	 * up to this point, then fail so that the superblock avoids
1736 	 * being written out as clean.
1737 	 */
1738 	if (allerror) {
1739 		brelse(sbbp);
1740 		return (allerror);
1741 	}
1742 	bp = sbbp;
1743 	if (fs->fs_magic == FS_UFS1_MAGIC && fs->fs_sblockloc != SBLOCK_UFS1 &&
1744 	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1745 		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1746 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS1);
1747 		fs->fs_sblockloc = SBLOCK_UFS1;
1748 	}
1749 	if (fs->fs_magic == FS_UFS2_MAGIC && fs->fs_sblockloc != SBLOCK_UFS2 &&
1750 	    (fs->fs_flags & FS_FLAGS_UPDATED) == 0) {
1751 		printf("%s: correcting fs_sblockloc from %jd to %d\n",
1752 		    fs->fs_fsmnt, fs->fs_sblockloc, SBLOCK_UFS2);
1753 		fs->fs_sblockloc = SBLOCK_UFS2;
1754 	}
1755 	fs->fs_fmod = 0;
1756 	fs->fs_time = time_second;
1757 	if (fs->fs_flags & FS_DOSOFTDEP)
1758 		softdep_setup_sbupdate(mp, (struct fs *)bp->b_data, bp);
1759 	bcopy((caddr_t)fs, bp->b_data, (u_int)fs->fs_sbsize);
1760 	ffs_oldfscompat_write((struct fs *)bp->b_data, mp);
1761 	if (suspended)
1762 		bp->b_flags |= B_VALIDSUSPWRT;
1763 	if (waitfor != MNT_WAIT)
1764 		bawrite(bp);
1765 	else if ((error = bwrite(bp)) != 0)
1766 		allerror = error;
1767 	return (allerror);
1768 }
1769 
1770 static int
1771 ffs_extattrctl(struct mount *mp, int cmd, struct vnode *filename_vp,
1772 	int attrnamespace, const char *attrname)
1773 {
1774 
1775 #ifdef UFS_EXTATTR
1776 	return (ufs_extattrctl(mp, cmd, filename_vp, attrnamespace,
1777 	    attrname));
1778 #else
1779 	return (vfs_stdextattrctl(mp, cmd, filename_vp, attrnamespace,
1780 	    attrname));
1781 #endif
1782 }
1783 
1784 static void
1785 ffs_ifree(struct ufsmount *ump, struct inode *ip)
1786 {
1787 
1788 	if (ump->um_fstype == UFS1 && ip->i_din1 != NULL)
1789 		uma_zfree(uma_ufs1, ip->i_din1);
1790 	else if (ip->i_din2 != NULL)
1791 		uma_zfree(uma_ufs2, ip->i_din2);
1792 	uma_zfree(uma_inode, ip);
1793 }
1794 
1795 static int dobkgrdwrite = 1;
1796 SYSCTL_INT(_debug, OID_AUTO, dobkgrdwrite, CTLFLAG_RW, &dobkgrdwrite, 0,
1797     "Do background writes (honoring the BV_BKGRDWRITE flag)?");
1798 
1799 /*
1800  * Complete a background write started from bwrite.
1801  */
1802 static void
1803 ffs_backgroundwritedone(struct buf *bp)
1804 {
1805 	struct bufobj *bufobj;
1806 	struct buf *origbp;
1807 
1808 	/*
1809 	 * Find the original buffer that we are writing.
1810 	 */
1811 	bufobj = bp->b_bufobj;
1812 	BO_LOCK(bufobj);
1813 	if ((origbp = gbincore(bp->b_bufobj, bp->b_lblkno)) == NULL)
1814 		panic("backgroundwritedone: lost buffer");
1815 	/* Grab an extra reference to be dropped by the bufdone() below. */
1816 	bufobj_wrefl(bufobj);
1817 	BO_UNLOCK(bufobj);
1818 	/*
1819 	 * Process dependencies then return any unfinished ones.
1820 	 */
1821 	if (!LIST_EMPTY(&bp->b_dep))
1822 		buf_complete(bp);
1823 #ifdef SOFTUPDATES
1824 	if (!LIST_EMPTY(&bp->b_dep))
1825 		softdep_move_dependencies(bp, origbp);
1826 #endif
1827 	/*
1828 	 * This buffer is marked B_NOCACHE so when it is released
1829 	 * by biodone it will be tossed.
1830 	 */
1831 	bp->b_flags |= B_NOCACHE;
1832 	bp->b_flags &= ~B_CACHE;
1833 	bufdone(bp);
1834 	BO_LOCK(bufobj);
1835 	/*
1836 	 * Clear the BV_BKGRDINPROG flag in the original buffer
1837 	 * and awaken it if it is waiting for the write to complete.
1838 	 * If BV_BKGRDINPROG is not set in the original buffer it must
1839 	 * have been released and re-instantiated - which is not legal.
1840 	 */
1841 	KASSERT((origbp->b_vflags & BV_BKGRDINPROG),
1842 	    ("backgroundwritedone: lost buffer2"));
1843 	origbp->b_vflags &= ~BV_BKGRDINPROG;
1844 	if (origbp->b_vflags & BV_BKGRDWAIT) {
1845 		origbp->b_vflags &= ~BV_BKGRDWAIT;
1846 		wakeup(&origbp->b_xflags);
1847 	}
1848 	BO_UNLOCK(bufobj);
1849 }
1850 
1851 
1852 /*
1853  * Write, release buffer on completion.  (Done by iodone
1854  * if async).  Do not bother writing anything if the buffer
1855  * is invalid.
1856  *
1857  * Note that we set B_CACHE here, indicating that buffer is
1858  * fully valid and thus cacheable.  This is true even of NFS
1859  * now so we set it generally.  This could be set either here
1860  * or in biodone() since the I/O is synchronous.  We put it
1861  * here.
1862  */
1863 static int
1864 ffs_bufwrite(struct buf *bp)
1865 {
1866 	int oldflags, s;
1867 	struct buf *newbp;
1868 
1869 	CTR3(KTR_BUF, "bufwrite(%p) vp %p flags %X", bp, bp->b_vp, bp->b_flags);
1870 	if (bp->b_flags & B_INVAL) {
1871 		brelse(bp);
1872 		return (0);
1873 	}
1874 
1875 	oldflags = bp->b_flags;
1876 
1877 	if (!BUF_ISLOCKED(bp))
1878 		panic("bufwrite: buffer is not busy???");
1879 	s = splbio();
1880 	/*
1881 	 * If a background write is already in progress, delay
1882 	 * writing this block if it is asynchronous. Otherwise
1883 	 * wait for the background write to complete.
1884 	 */
1885 	BO_LOCK(bp->b_bufobj);
1886 	if (bp->b_vflags & BV_BKGRDINPROG) {
1887 		if (bp->b_flags & B_ASYNC) {
1888 			BO_UNLOCK(bp->b_bufobj);
1889 			splx(s);
1890 			bdwrite(bp);
1891 			return (0);
1892 		}
1893 		bp->b_vflags |= BV_BKGRDWAIT;
1894 		msleep(&bp->b_xflags, BO_MTX(bp->b_bufobj), PRIBIO, "bwrbg", 0);
1895 		if (bp->b_vflags & BV_BKGRDINPROG)
1896 			panic("bufwrite: still writing");
1897 	}
1898 	BO_UNLOCK(bp->b_bufobj);
1899 
1900 	/*
1901 	 * If this buffer is marked for background writing and we
1902 	 * do not have to wait for it, make a copy and write the
1903 	 * copy so as to leave this buffer ready for further use.
1904 	 *
1905 	 * This optimization eats a lot of memory.  If we have a page
1906 	 * or buffer shortfall we can't do it.
1907 	 */
1908 	if (dobkgrdwrite && (bp->b_xflags & BX_BKGRDWRITE) &&
1909 	    (bp->b_flags & B_ASYNC) &&
1910 	    !vm_page_count_severe() &&
1911 	    !buf_dirty_count_severe()) {
1912 		KASSERT(bp->b_iodone == NULL,
1913 		    ("bufwrite: needs chained iodone (%p)", bp->b_iodone));
1914 
1915 		/* get a new block */
1916 		newbp = geteblk(bp->b_bufsize, GB_NOWAIT_BD);
1917 		if (newbp == NULL)
1918 			goto normal_write;
1919 
1920 		/*
1921 		 * set it to be identical to the old block.  We have to
1922 		 * set b_lblkno and BKGRDMARKER before calling bgetvp()
1923 		 * to avoid confusing the splay tree and gbincore().
1924 		 */
1925 		memcpy(newbp->b_data, bp->b_data, bp->b_bufsize);
1926 		newbp->b_lblkno = bp->b_lblkno;
1927 		newbp->b_xflags |= BX_BKGRDMARKER;
1928 		BO_LOCK(bp->b_bufobj);
1929 		bp->b_vflags |= BV_BKGRDINPROG;
1930 		bgetvp(bp->b_vp, newbp);
1931 		BO_UNLOCK(bp->b_bufobj);
1932 		newbp->b_bufobj = &bp->b_vp->v_bufobj;
1933 		newbp->b_blkno = bp->b_blkno;
1934 		newbp->b_offset = bp->b_offset;
1935 		newbp->b_iodone = ffs_backgroundwritedone;
1936 		newbp->b_flags |= B_ASYNC;
1937 		newbp->b_flags &= ~B_INVAL;
1938 
1939 #ifdef SOFTUPDATES
1940 		/*
1941 		 * Move over the dependencies.  If there are rollbacks,
1942 		 * leave the parent buffer dirtied as it will need to
1943 		 * be written again.
1944 		 */
1945 		if (LIST_EMPTY(&bp->b_dep) ||
1946 		    softdep_move_dependencies(bp, newbp) == 0)
1947 			bundirty(bp);
1948 #else
1949 		bundirty(bp);
1950 #endif
1951 
1952 		/*
1953 		 * Initiate write on the copy, release the original to
1954 		 * the B_LOCKED queue so that it cannot go away until
1955 		 * the background write completes. If not locked it could go
1956 		 * away and then be reconstituted while it was being written.
1957 		 * If the reconstituted buffer were written, we could end up
1958 		 * with two background copies being written at the same time.
1959 		 */
1960 		bqrelse(bp);
1961 		bp = newbp;
1962 	} else
1963 		/* Mark the buffer clean */
1964 		bundirty(bp);
1965 
1966 
1967 	/* Let the normal bufwrite do the rest for us */
1968 normal_write:
1969 	return (bufwrite(bp));
1970 }
1971 
1972 
1973 static void
1974 ffs_geom_strategy(struct bufobj *bo, struct buf *bp)
1975 {
1976 	struct vnode *vp;
1977 	int error;
1978 	struct buf *tbp;
1979 	int nocopy;
1980 
1981 	vp = bo->__bo_vnode;
1982 	if (bp->b_iocmd == BIO_WRITE) {
1983 		if ((bp->b_flags & B_VALIDSUSPWRT) == 0 &&
1984 		    bp->b_vp != NULL && bp->b_vp->v_mount != NULL &&
1985 		    (bp->b_vp->v_mount->mnt_kern_flag & MNTK_SUSPENDED) != 0)
1986 			panic("ffs_geom_strategy: bad I/O");
1987 		nocopy = bp->b_flags & B_NOCOPY;
1988 		bp->b_flags &= ~(B_VALIDSUSPWRT | B_NOCOPY);
1989 		if ((vp->v_vflag & VV_COPYONWRITE) && nocopy == 0 &&
1990 		    vp->v_rdev->si_snapdata != NULL) {
1991 			if ((bp->b_flags & B_CLUSTER) != 0) {
1992 				runningbufwakeup(bp);
1993 				TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
1994 					      b_cluster.cluster_entry) {
1995 					error = ffs_copyonwrite(vp, tbp);
1996 					if (error != 0 &&
1997 					    error != EOPNOTSUPP) {
1998 						bp->b_error = error;
1999 						bp->b_ioflags |= BIO_ERROR;
2000 						bufdone(bp);
2001 						return;
2002 					}
2003 				}
2004 				bp->b_runningbufspace = bp->b_bufsize;
2005 				atomic_add_long(&runningbufspace,
2006 					       bp->b_runningbufspace);
2007 			} else {
2008 				error = ffs_copyonwrite(vp, bp);
2009 				if (error != 0 && error != EOPNOTSUPP) {
2010 					bp->b_error = error;
2011 					bp->b_ioflags |= BIO_ERROR;
2012 					bufdone(bp);
2013 					return;
2014 				}
2015 			}
2016 		}
2017 #ifdef SOFTUPDATES
2018 		if ((bp->b_flags & B_CLUSTER) != 0) {
2019 			TAILQ_FOREACH(tbp, &bp->b_cluster.cluster_head,
2020 				      b_cluster.cluster_entry) {
2021 				if (!LIST_EMPTY(&tbp->b_dep))
2022 					buf_start(tbp);
2023 			}
2024 		} else {
2025 			if (!LIST_EMPTY(&bp->b_dep))
2026 				buf_start(bp);
2027 		}
2028 
2029 #endif
2030 	}
2031 	g_vfs_strategy(bo, bp);
2032 }
2033 
2034 #ifdef	DDB
2035 
2036 static void
2037 db_print_ffs(struct ufsmount *ump)
2038 {
2039 	db_printf("mp %p %s devvp %p fs %p su_wl %d su_deps %d su_req %d\n",
2040 	    ump->um_mountp, ump->um_mountp->mnt_stat.f_mntonname,
2041 	    ump->um_devvp, ump->um_fs, ump->softdep_on_worklist,
2042 	    ump->softdep_deps, ump->softdep_req);
2043 }
2044 
2045 DB_SHOW_COMMAND(ffs, db_show_ffs)
2046 {
2047 	struct mount *mp;
2048 	struct ufsmount *ump;
2049 
2050 	if (have_addr) {
2051 		ump = VFSTOUFS((struct mount *)addr);
2052 		db_print_ffs(ump);
2053 		return;
2054 	}
2055 
2056 	TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2057 		if (!strcmp(mp->mnt_stat.f_fstypename, ufs_vfsconf.vfc_name))
2058 			db_print_ffs(VFSTOUFS(mp));
2059 	}
2060 }
2061 
2062 #endif	/* DDB */
2063